Latest / The BioHub - by Avetix / Episode 158 - Daniel Hader - Model Medicines

Transcript

Episode 158 - Daniel Hader - Model Medicines · The BioHub - by Avetix

0:06
Hey everyone and welcome to another episode of the Bio Hub
0:09
podcast. Today I'm joined by Daniel
0:12
Haders, Co founder and CEO of Model Medicines.
0:16
Daniel has a background in biomedical engineering and has
0:18
spent his career working across the interface of computation
0:22
biology and drug development. At Model Medicines, he's
0:25
building an AI driven drug discovery platform that
0:28
deliberately takes a different approach to much of the
0:31
industry, focusing on applying machine learning where humans
0:35
have historically failed, rethinking how much models are
0:38
built, scaled to drive real outcomes, and really looking
0:42
forward to unpacking both the science and some other thinking
0:44
behind that approach today. Daniel, welcome to the podcast.
0:48
Thank you for having me, it's a true honor to be here.
0:50
Amazing. And I thought we could start for
0:53
our listeners, a bit of a background into yourself.
0:55
And you know what really LED you to start Model medicines.
0:59
Sure. I, I appreciate that
1:01
opportunity, the, the way I think about my journey and, and
1:06
I say I, I grew up a long time ago, I'll say in, in, in the
1:13
faraway Galaxy of, of Kentucky. And I was early on as a as a
1:20
kid, I was, I was always interested in math, chemistry
1:25
and, and biology. And a little school named
1:29
Columbia took pity on me and invited me to join their class
1:36
in 1997. And while I was at Columbia, I,
1:41
I knew I wanted to work at the interface of multiple
1:46
disciplines that I perceived, at least at that time, were on the
1:51
precipice of, of changing the world for, for human beings in a
1:56
very meaningful way. And, and while I was there, I
2:00
was given the opportunity to, to learn at the intersection of, of
2:06
physics and chemistry and biology and computer science and
2:11
data science and, and mathematics while earning this,
2:16
this new degree in this field called biomedical engineering.
2:20
And, and that formative training that, that I had in the late 90s
2:25
at the intersections of those disciplines.
2:28
I didn't know it at the time, but ultimately set me up to
2:33
become the CEO of, of model Medicines because what, what
2:39
we've seen now is that the truly miraculous discoveries, the
2:44
things that change human lives for, for all of us globally
2:49
occur at these intersections of, of disciplines.
2:54
And so I was, I was very lucky early on to, to work at that, at
2:58
that intersection between like I said, data science, computer
3:02
science, chemistry, biology, mathematics and, and whatnot.
3:07
And, and that set me on a pathway where I was lucky enough
3:11
as a graduate student to have a National Science Foundation
3:16
fellowship and work with Johnson and Johnson on a program where
3:22
we use those disciplines to identify, using computer science
3:27
and data science to identify very niche and specialized
3:32
chemical space. And then spend time in the lab
3:35
doing novel chemical synthesis. And then spent time in the
3:39
biology department validating those discoveries using cell
3:43
culture, PCR, electron microscopy, and everything under
3:48
the sun. And, and so that took my, my
3:50
undergraduate training and turned it into something that
3:54
was tangible in this case, patents and, and papers.
3:58
And then I was lucky enough to then work in early stage drug
4:02
development in a venture back pharmaceutical company.
4:05
And all of a sudden, what then layered on top of it was not
4:09
just merging scientific disciplines, but merging
4:12
scientific disciplines with patent law and corporate
4:17
development and venture capital financing.
4:20
And that really, really stoked my my interest as someone who
4:26
didn't just want to discover things, but wanted to discover
4:29
things that had practical applicability for huge swaths
4:33
of, of the human population. And so while I was there, I was
4:37
lucky enough to work on a drug program that we took from
4:41
concept to the clinic. And ultimately that program that
4:46
I worked on with one of my other cofounders here at Model
4:49
Medicine, Sean Russell, ultimately morphed into an
4:52
approved drug called Sequa, which was approved in the 20
4:56
teens. And and from that moment on,
4:58
I've constantly been working in the venture backed healthcare
5:02
space, whether that has been in the intersection of material
5:07
science and chemistry or at the intersection of data science,
5:11
broadband, wireless medical devices, or even more recently
5:17
with one of the other members of the team funding as a venture
5:21
capitalist digital health companies spanning from
5:25
cybersecurity and model cybersecurity and medical
5:28
devices to the world's first VR therapeutic approved by the FDA.
5:34
And while it was there on, on this journey, basically, I
5:39
realized that the infrastructure that was available to scientists
5:44
and entrepreneurs, the infrastructure that we could
5:47
only have dreamed about in the late 90s when I was starting my
5:51
scientific career, now was available to everyone, right?
5:55
The, the ability to have compute at a moment's notice.
6:00
And this changed what was possible when I was a young
6:03
graduate student that I was mentioning before, we use
6:06
Silicon Graphics computers plugged into a wall locally.
6:10
And if the power went out in the lab, you had to restart
6:12
everything and it was a one week delay in your research.
6:16
But now we have CPUs and GPUs available at a moment's notice
6:22
in huge quantities, compute power that we can never have
6:25
imagined would have been possible 25 years ago.
6:28
And in between this background that I had in computer science,
6:34
data science, chemistry, biology, mathematics, taken
6:37
together with 20 years of experience in drug development,
6:42
device development and investing in AI or machine learning
6:48
enabled healthcare companies. We arrived at a moment with
6:53
another early member of a team, Patrick O'Neill.
6:56
And, and we looked out at the audience and we decided or
6:59
looked out in the world, I should say the audience, we
7:01
looked out in the world and we said we really think this idea
7:05
of AI drug development has reached an inflection point
7:09
because of the maturity of the scientific disciplines together
7:13
with the maturity of AI and the infrastructure to power real AI.
7:21
And based on that, we set out to 1st invest in an AI driven
7:28
pharmaceutical company. And what we did was we built a a
7:34
relatively extensive and detailed set of underwriting
7:39
standards for that investment in AI driven drug discovery.
7:45
And probably no surprise to you that included that we believed
7:50
that you needed to have data science, machine learning, AI,
7:55
chemistry, biology, mathematics, sitting next to people that had
8:01
developed drugs before, sitting next to people that had financed
8:05
deep tech venture backed companies before as well.
8:09
And it developed them. And unfortunately we went out to
8:12
the market and there's a lot of great companies in this space.
8:15
Let me first say that folks that have pioneered the space, the
8:19
Schrodinger's, the recursions, the insilicos of of the world.
8:24
But when we were looking for the next company, we, we didn't
8:28
necessarily see a place where all of those disciplines were
8:32
merging into one. We saw places where computer
8:37
scientists and data scientists were sitting with one another.
8:40
We saw places where chemists and biologists were sitting with one
8:44
another. But we didn't see a place where
8:47
they were all sitting next to each other on one team together
8:50
with people that had developed drugs before, together with
8:54
people that had financed venture backed deep tech companies
8:58
before. And, and ultimately because of
9:01
that, that we didn't find a company that fit our
9:03
underwriting standards. And so we decided to build it
9:07
first as a research project. And then once we had built the
9:11
software, the early versions of the software and made our first
9:14
discoveries, we brought it to life and fully funded the
9:18
company. And the manifestation of those
9:22
underwriting standards that I mentioned before is, is Model
9:25
Medicines. And so it's, it's like the old
9:28
story, you know, Model medicines is now we are now publishing
9:33
data that's one of one data. But in reality, the model
9:37
Medicine story is a story that goes back to probably, you know,
9:41
Kentucky, Covington, KY in 1996 before I ever entered Columbia.
9:47
And, and it's been a 30 plus year journey to get here.
9:51
And all the steps in those journey were required to have
9:55
the perspective that we needed to create a company that now
9:59
appears to be doing over and over one of one things in not
10:06
only AI space, but also in the biotech or pharma space as well.
10:10
Amazing. I think that's a great overview
10:12
and I think it brings us on to our next question of what is
10:17
model medicines doing differently to the majority of
10:21
AI drug discovery companies at the moment.
10:23
Yeah. And I, I really appreciate that
10:26
that question. And let let me first say that I
10:31
believe that model medicines that there are multiple pathways
10:35
to arrive at the same answer, or in this case, an answer being a
10:42
drug that fundamentally can change the the health span, the
10:48
lifespan for a, a group of patients right there.
10:53
There's no only one way to do it.
10:56
But I, I will say that we, when we researched the AI driven drug
11:01
discovery industry, we walked away with in general, an idea
11:08
that we thought it could be done essentially the opposite way of
11:16
the way that it was being done dogmatically across most, if not
11:22
all companies in the space or most or not all academic
11:27
approaches. And So what do I mean by that,
11:30
right? To get away from the the
11:32
hyperbole of that answer and to answer you specifically what
11:37
what I mean is that when we looked at the technology
11:41
underlying AI driven drug discovery, we saw that in
11:45
general, there was this focus on gigantic data sets to train your
11:51
models. And there was focus this idea of
11:55
if I have more data, I can train better models and I can discover
12:00
new drugs. And we, to be completely honest,
12:04
didn't understand that approach because my teaching when it
12:11
comes to data science and computer science was that more
12:16
data. Well, let me take a step back.
12:19
We are in an industry where we're interested in discovering
12:22
one of one drugs, right? It's a what people might call a
12:25
power law rule. So we're interested in
12:28
discovering outliers. The problem is if you have lots
12:33
and lots of data, what typically happens is all that data is
12:38
concentrated in the same area of your data plot.
12:44
Your in this case, say your chemical space.
12:48
And the problem is, is when you have overlapping data.
12:51
I don't want to discover what's in my training set.
12:55
I want to discover something that is completely separate and
12:59
novel to my training set. But if I have an overwhelming
13:03
amount of data in areas of chemical space that I already
13:07
understand, those data points swamp out a model's ability to
13:13
learn and discover the power law result that I want.
13:18
The one of one result, the outlier.
13:22
And so rather than focusing on having a lot of data that was
13:26
concentrated, we said we wanted to train our models on a very
13:32
small amount of data that was extremely diverse, right?
13:38
And so this was, this was going and we wanted to have models
13:42
therefore that wouldn't overfit and would find those outliers.
13:46
So this idea of other folks train their models often times
13:50
on millions or billions of data points, we train our models on a
13:55
couple thousand data points, right?
13:58
So that's one of the big things that separates us.
14:02
Now. What it what does that mean?
14:04
So we ended up by training our models on these diverse but very
14:09
small training sets. When we did retrospective
14:12
validation of our models, we achieved area under the curve
14:17
scores that are highly predictive for drug discovery
14:21
but are lower than area under the curve scores that you might
14:25
find that we get you promoted in academia, right.
14:28
So our area under the curve scores are predictive, but
14:31
they're not approaching one. But our idea was that
14:35
retrospective modeling has nothing to do with discovering
14:39
one of one drugs in the real world, right?
14:42
It's just a baseline idea. So we created models that
14:47
performed more poorly in retrospective modeling, but we
14:51
believed we're ultimately more creative for discovering truly
14:55
new knowledge and truly new novel drugs that could change
15:00
the face of healthcare for patients globally.
15:04
Once we had to discover these creative models and built them,
15:08
then the problem becomes your model can't discover chemistry
15:13
to solve biology that it hasn't seen before.
15:17
So we believed that the issue was throughput.
15:21
And this is where we we believe scale was important, right?
15:24
So other people believe scale and training was the most
15:27
important. We believe that scale in
15:30
inference or throughput was the important scaling factor that
15:36
you needed. And what do I mean there?
15:38
So all of the drugs that are possible have been estimated to
15:42
be around 10 to the 60th, right? All of the drugs that have been
15:48
in humans is somewhere around 10 or 15,000, right?
15:54
And once you once you absorb that idea, what you realize is
15:58
that our knowledge of medicinal chemistry as applied to biology
16:04
is literally less than a single drop of water in all the oceans
16:08
of the world, right? And once you accept that fact
16:13
that we know essentially nothing, then the goal has to be
16:17
to get beyond the bounds of anything that we know.
16:21
And So what we believed was that our model needed to see that 10
16:26
to the 60th chemistry or something approaching it.
16:30
And so if we built predictive models that were predictive in
16:33
the real world, as I was mentioning before, and then we
16:36
allowed them to do more inference or see more chemistry
16:41
than any other model on Earth, then we would be able to
16:45
discover drugs that to this point had been undiscoverable or
16:50
been impossible to discover. Yeah, right.
16:53
So this focus on small training sets to create creative models
16:58
that do very well on predicting unseen chemistry and then
17:02
showing them more, showing those models more chemistry than any
17:06
model I've ever seen before. But that's the technical aspect,
17:09
right? Then there's the biological or,
17:12
or the biotech aspect of like how do you apply those technical
17:17
concepts? And for us, what we decided to
17:21
do is that we believe that human medicinal chemist and biologist
17:26
are, look, we have amazing medicines, right, that are out
17:30
there that save millions of lives every year, change health
17:33
spans, right? Change, change families forever
17:36
because they save people's lives.
17:38
And we have done really well as humans on some very limited
17:42
types of drugs and diseases and targets.
17:46
We thought the point of AI was to apply this tool to places
17:51
specifically where there was a there was a grail drug, where a
17:56
grail target or a grail product that humans knew have known
18:02
needed to exist, but where humans had attempted to solve
18:07
that problem and find that grail solution, but had failed
18:12
definitively and obviously. And what our idea was, is that
18:17
we should take AI and only apply it to problems that were
18:22
impossible for humans to solve. And then within that area of of
18:27
problems where humans had failed and we started to call the
18:32
solutions undruggable or impossible.
18:36
Within that, we decided to focus on a very specific set of
18:41
biological targets. So we wanted to focus on either
18:46
discovering what we call biological choke points or
18:49
druggy, known but undruggable, quote UN quote biological choke
18:54
points. And what is a biological choke
18:56
point? A biological choke point is a
19:00
target that if you drug it is well understood it can solve 10s
19:07
to potentially hundreds of diseases with a single drug and
19:11
a single target. So these grail sort of solutions
19:15
where if I do one thing, I can change millions of lives, if not
19:20
10s of millions of lives overnight.
19:23
And so, so it was this idea, small data sets rather than
19:28
large creative models, rather than dogmatic models, huge
19:34
inference sets that gets us into huge inference sets that gets us
19:39
into deep chemical space. And then focusing this on the
19:43
places that are undruggable, the places where humans have failed,
19:47
the places where we say the IT is impossible to discover
19:52
something. And that's where we have focused
19:54
all of our work on model medicines, right?
19:57
Doing zigging while others are zagging from a technical point
20:00
of view and then only trying to solve the impossible problems
20:04
today. Yeah, that is really interesting
20:06
about how that that platform is working at a higher level.
20:09
And when you talk about the biological chalk points, when
20:13
you hit one of those, what does that mean in a biological
20:16
context? Yeah, it's another, another
20:20
great question. And we have, I think it's best,
20:23
I've been talking a lot about concepts, right.
20:27
And so I think maybe the best way at a certain point people
20:31
don't want to hear about concepts they want, they want to
20:33
know, do your concepts work. And we've reduced the technology
20:37
and this approach to drug discovery to practice in two,
20:42
two separate drug programs that if you give me a, a moment,
20:46
I'll, I'll describe both of them in in one case, there is a
20:51
target called BRD four and BRD 4.
20:56
The discovery of it and the understanding of the importance
21:00
of this target, which is a transcription factor was
21:04
pioneered about 2025 years ago. And this is a target that we
21:11
understand as pharmaceutical scientists that if you drug it,
21:16
you can solve diseases ranging from oncology indications to I
21:23
and I indications to cardiovascular indications, to
21:28
infectious disease indications, to even CNS and
21:31
neurodegenerative indications like Alzheimer's, right?
21:35
So one target and the world knows it's validated by some of
21:39
the most brilliant scientists on Earth that if you can drug this
21:43
target, you can potentially solve or aid in solving all of
21:49
those indications, right? And I didn't even get into
21:51
detailed indications. I was giving you categories,
21:54
right? And inside of each of those
21:55
categories is probably 10 indications with an oncology I
21:59
and I cardiovascular, etcetera, etcetera, right?
22:04
So, but what was the problem in this case?
22:06
It was a problem of drug selectivity.
22:09
Every drug that had been brought forward to clinical trials to
22:13
drug BRD 4 has also drugged A sibling protein or a brother
22:20
sister protein called BRD 3. And when you drug BRD three, you
22:26
get a dose limiting blood toxicity called
22:29
thrombocytopenia. Thrombocytopenia ends every
22:34
clinical trial on the spot and the problem has been, there's
22:39
been approximately about 20 compounds have entered the
22:42
clinic against BRD four and they have all failed due to this dose
22:48
limiting blood toxicity. And so the problem has been we
22:52
could drug BRD 4, but we've never been able to drug BRD 4
22:56
without drugging BRD 3. But we know that if we could do
23:00
that, the upside for humanity and health span and lifespan is
23:05
really hard to, to calculate. And, and the way that you know
23:10
that the Pharmaceutical industry understands the importance of
23:13
this target is because typically when a new target is pioneered,
23:18
a drug will be developed. And if it has a tox liability,
23:21
maybe one other company develops a drug against that and tries
23:25
again. And if it fails, people stop
23:28
developing in that space quite often.
23:30
In this case, a double digit number of companies, right?
23:34
It brought 20 drugs to bear in this space, which tells you by
23:39
definition how important this target is and how much we truly
23:43
want to create a solution for it.
23:46
At Model Medicines using that platform I described before with
23:51
small training sets, creative models and large inference, we
23:56
in early testing, early non clinical testing appear to have
24:02
discovered a drug that is potent, selective and novel in a
24:09
way that no drug has previously been potent, selective or novel
24:15
in this space. And so it's still now it's early
24:20
days. I would say we're in the middle
24:22
of non clinical development, but the early numbers tell us, you
24:28
know, that this validation, wet lab validation, say that this
24:34
drug by by the numbers, right, without hyperbole or without
24:38
hand waving reports potency, selectivity and novelty that is
24:44
not reported beyond anything that's reported in the
24:47
literature to date. And in the second instance,
24:50
because now you say, well, anybody can get lucky once,
24:54
right? And so, or a broken clock is,
24:57
you know, right, once every 12 hours, right as as they say.
25:01
So, So the question is, can you do it again?
25:03
And, and we see this in our infectious disease program.
25:08
So what motivated us there? And actually, this was our first
25:11
program was that we looked around in the early days of the
25:17
pandemic and we said, why are we in this mess that we are in?
25:21
Why is there no drug already approved, stockpiled it on the
25:26
shelves to deal with this novel threat.
25:30
And what we learned was that no, there is, to the best of our
25:35
knowledge, there's no drug that has ever been created for a
25:39
virus before that virus existed, right?
25:44
This, this idea of predictive drug discovery, it didn't,
25:48
wasn't something that existed. And therefore there was this
25:51
dogma. There's also this dogma that
25:54
every virus requires its own drug, right?
25:57
So you have sevastobir for hepatitis C, you have tenofovir
26:03
for hepatitis B, you have Tamiflu for influenza, you have
26:08
Patchlovid now for SARS COV 2. But the idea of having one drug
26:15
for all of those viruses is something that we were told was
26:19
impossible. And having one drug that didn't
26:23
hit a human target, but hit a viral target, we were told was
26:27
even more impossible. So this said everyone, what do
26:31
we want? Right?
26:31
So what everyone wants is a single oral pill that directly
26:36
treats all major viruses, Hep C, Hep B, Hep D, influenza, RSV and
26:44
coronaviruses, right? One pill for all of them.
26:48
Everyone wants that, and everyone believes that that is
26:51
impossible to have an oral medication that directly acts on
26:55
a virus that inhibits all of those viruses simultaneously.
27:01
We may be because we are naive, but also because the power of
27:06
AI. We said why isn't this possible?
27:10
And so leveraging our platform, we discovered a new target,
27:16
druggable target that's on the viral polymerase that is
27:20
conserved across viral biology. And then we discovered a drug
27:26
for that target. And in late non clinical testing
27:31
we have demonstrated that the drug works directly against
27:35
coronaviruses, multiple coronaviruses, influenza, RSV,
27:41
Hep C, Hep B and Hep D, something that we were told was
27:46
was impossible. And, and again, the reason why
27:50
we were able to discover it is because first we eschewed dogma
27:55
that was there and said why not? And then we simply got to work.
27:59
And we, we separated what people believe and what people know.
28:03
And then we let the models guide us rather than us telling the
28:08
models where to go. We trained the models, provided
28:11
of course some parameters, but then had the models drive us and
28:16
the models drove us to the target and to the drug to a data
28:22
set that we cannot find another instance in the literature,
28:27
whether preclinically, clinically or anywhere where a
28:30
single oral medication that has definitive mechanism of action
28:35
against the virus directly has demonstrated this breath of, of
28:43
activity. And then what's important as
28:45
well as not only is it active, but doesn't work as well or
28:48
better than the standards of care.
28:50
And so to date, and this is a growing list to date, we've
28:54
demonstrated either equivalency or superiority to Remdissevier,
29:00
Paxlovid and Savassevier, which is also known as Savaldi and a
29:06
couple other forms as well. So that program is getting ready
29:10
for an Ind submission late this year and we're looking to be in
29:13
the clinic early next year with that program.
29:16
But these are two cases where the world said something was
29:20
impossible and we pushed that idea to the side, applied the
29:26
technology that I described before, and discovered something
29:29
that people said was was impossible.
29:32
And so we now become even more convinced that there is this
29:37
separation between, again, what people think and what people
29:41
know in this space. And we're really excited to keep
29:45
on pushing the boundaries and redefine what we know and really
29:50
bring these life changing solutions to huge swaths of the
29:55
human population globally. Yeah.
29:57
And that's just such an interesting way to do so.
30:00
And it sounds like you've focused on problems that humans
30:04
have struggled to solve. But how do you go about choosing
30:07
which problems to work on? This is a this is a great
30:11
question and and you know, to be to be completely honest, there
30:16
are there are multiple sides to it.
30:18
And it is when, when you have a tool that can go in any
30:22
direction going when, when everything is possible, choosing
30:29
where to go first is a much harder decision than, than you
30:34
might think. Because in, in reality, as a, as
30:37
a human, as a, as a humanist, we all have family, right?
30:42
We have, we have peers and friends that are our age.
30:45
You know, we have younger folks, either your children or your
30:49
nieces and nephews. We have elderly parents and
30:53
aunts and uncles. And, and then we look out into
30:56
the world, right? And you see, everyone is a
30:59
patient at one point in their life and everyone needs help.
31:04
And there are so many diseases that are so awful that need a
31:09
need a solution, right? And so how do you pick what
31:13
comes first? And, and in many ways, what this
31:18
LED us to this idea of focusing on the biological choke points,
31:23
because we wanted to focus on places where I didn't have to
31:27
split hairs, where where one discovery could solve 20
31:32
different diseases, 1 discovery could help millions or hundreds
31:37
of millions of people at one time with that, with that core
31:40
discovery. So what has guided us to this
31:43
point is that human perspective. Let's find the place where one
31:47
solution solves the most number of diseases.
31:51
And then in addition to that, that, you know, the cold truth
31:54
about being in, in the pharmaceutical sciences is you
31:59
also have to be aware of where the those who have accumulated
32:05
investable capital, what are the trends at the time that are
32:11
motivating them? And where do they see the
32:13
greatest returns? Because, you know, the best
32:16
scientists will, will understand that there is no discovery
32:22
without the financial partners to support those scientific
32:26
discoveries. And we may want something, but
32:29
if, if it's, if it's not financeable, if you can't
32:32
finance that solution, it doesn't matter if you're right
32:36
at the scientist as a scientist. And so we've merged this idea to
32:41
solve the greatest number of diseases in the shortest period
32:44
of time with the pressures that we understand that our financing
32:48
partners are under to deliver in certain areas.
32:51
And when we take that scientific approach together with that
32:55
clinical approach, together with that financial approach that
32:59
ultimately allows us to arrive at what we do 1st and what we do
33:03
second. But but I want to be clear, the
33:05
way that we think about this is not, this is what we're going to
33:09
do and this is not what we're going to do.
33:11
We think about this is that this is the first thing because it
33:14
works scientifically, it has huge unmet medical need and it
33:18
you can finance this program and then we have a second one and
33:22
then the third one and then a fourth one.
33:24
But in reality, the 50th one on our list is no more or less
33:28
important than the first one on our list.
33:30
There's just this combined, this combination of factors, like I
33:34
said, between science, between clinical development and
33:38
financing that ultimately leads you to certain places.
33:42
Amazing. And you mentioned earlier about
33:44
this AI driven hub and Spark model.
33:48
How does that work in practice? Yeah.
33:51
And, and this again, this is such a good question because
33:55
when, when you have a technology that can do so much, so fast,
33:59
you're one of the problems that you have is making sure the
34:05
financing is there so that the science doesn't slow down once
34:10
the science is validated, right? And, and one of the, the things
34:15
that, that people fully understand now is we use AI to
34:19
go from concept to the clinic. And we can do that very quickly
34:24
in a very, very capital efficient way.
34:28
And we can discover drugs with a very high probability of
34:33
clinical success where we've solved numerous clinical
34:37
problems a priori before they become clinical problems and
34:42
shut down clinical trial programs.
34:44
But nonetheless, you still have to run clinical trials and
34:48
clinical trials are expensive, right?
34:50
There's just no way around it. And they take a long time
34:54
because patient recruitment is hard, right?
34:57
And so if, if you realize that now for, for us, the discovery
35:04
into the clinic is fast and easy.
35:06
The roadblock then becomes the, the pace of patient recruitment
35:11
and simply the time and cost of the clinical trials, right.
35:15
And so now what your problem is, is really it's a financial
35:18
problem as we were discussing before that we have more drugs
35:23
than there is money, you know, in in the world to to be
35:27
completely honest, right? And So what you do is you have
35:30
to be creative on ways to most efficiently finance the largest
35:36
number of clinical trials so that you can get the most drugs
35:39
to the most patients in the shortest period of time.
35:43
And one of the things that our peers at Schrodinger and others
35:48
and well have a pioneer places like Roy, you know Roy Van does
35:55
a version of this and some of these, there are several other
35:57
companies bridge bio that have hub and spoke models.
36:01
But the idea of a hub and spoke model is that it allows you to
36:06
access all the different pools of capital that are available in
36:10
the world, right? So one pool of capital is
36:13
private capital. And so sometimes private capital
36:18
like certain things and not other things.
36:20
So you create products, programs and then turn you can turn them
36:25
into companies and then those companies can be financed by the
36:29
people that are particularly interested in that particular
36:32
program. And what I mean is that you as
36:36
an investor might not be interested in AI, so you don't
36:40
want to invest in the core model medicines business, but you are
36:45
very bullish on BRD 4 or very bullish on infectious disease.
36:51
And so if you have a hub and spoke model, you can take a
36:54
program, put it into a subsidiary entity and then
36:59
finance that with the people who are passionate in particular
37:04
about that entity. So the people that are
37:07
interested deeply in AI in our core business invest in equity
37:12
and model medicines. Those that are deeply interested
37:15
in virology could invest in a a spoke company that is specific
37:21
to to virology. Those focused on the on the
37:24
potential BRD 4 can do it the same way.
37:27
Now as I mentioned, you can, you can have this hum and spoke and
37:31
do it just say with private capital, because private capital
37:34
has silos that like different things.
37:36
But when you also do the hub and spoke, what it allows you to do
37:40
is potentially access public capital as well in addition to
37:44
the private capital. And then public capital can be
37:48
equity capital or it can be debt capital, right?
37:52
And in all of these things together, what you're really
37:55
doing is you're, you're building corporate entities that are fit
38:01
for various silos of capital and the type of investments that
38:06
that capital wants to make, right?
38:08
So we think about there's public equity, there's private public
38:12
equity, public debt, private private equity capital, private
38:17
credit, so on and so forth. Each of those you can create
38:22
companies and corporate structures for them.
38:26
And then within those various groups, there are groups that
38:30
are interested in AI, some in virology, some in BRD 4.
38:34
And so the idea is reducing the friction via hub and spoke
38:39
corporate structures so that you can match investors with the
38:45
technologies or the therapeutics that they're most interested in
38:49
investing in and reducing the friction between investments for
38:57
those various groups. And I'll give you a few
38:59
examples. But the type of scientist or
39:03
entrepreneur that that I would like to have dinner with and,
39:06
and then the type maybe not surprising is the people that
39:12
have the scientists that were at one point considered outside of
39:20
outside of the conventional world of science.
39:24
And then maybe they struggled and they believed in their
39:29
science and they kept on pushing that rock up up the hill.
39:33
And then one day they went from being an outsider with ideas
39:38
that seemed crazy to being the insider that the textbooks were,
39:43
were written about, right. And, and so I, I think, you
39:48
know, one of my, one of my favorite quotes, and this is one
39:50
of the people that I would love sort of past that I would love
39:54
to have dinner with is, is Galileo has this quote and he
39:58
says that all truths are easy to understand once they are
40:02
discovered. The point is to discover them,
40:05
right? And I, I love that quote because
40:09
what it what it means is that there are.
40:14
After something is discovered, and especially as the best
40:17
scientists have taken a complex idea and made it simple,
40:22
everyone always says, well that's obvious.
40:25
But typically those scientists that have changed the world,
40:29
there was a moment where they were people thought that what
40:33
they were talking about was impossible or their approach
40:37
made no sense. And then one day something
40:40
happens and they go from, oh, that's impossible to yeah, it's
40:45
obvious. And anybody could have
40:48
discovered this as well. And in Galileo from from
40:52
somebody from the deep pass, Galileo represents this, right?
40:55
This is somebody who followed the data, believed in the
41:00
science, even at the risk of being excommunicated from the
41:04
church, right? And he literally risked
41:07
everything because he was convinced.
41:10
He followed the data and was convinced of the righteousness
41:13
of of the data. And in a more modern place,
41:17
somebody that that I similarly really admire is Doctor Caitlin
41:22
Carico, who was formula of Upenn and also a Bioentech.
41:27
And, and I think with Doctor Carico, what you see here is
41:31
that now she appears as someone who is in her 70s that has won a
41:37
Nobel Prize and is was far ahead of the curve.
41:41
But there was a moment in time where she couldn't get tenure at
41:47
Upenn, right? She couldn't get funding from
41:51
any of the agencies that typically would fund her
41:54
research in the United States, right?
41:57
And this is somebody who went from that to a Nobel Prize and
42:02
all of the work was the same, right?
42:05
She's the same scientist, right? That she was when she couldn't
42:09
get tenure and as a Nobel laureate, right?
42:12
She had the same ideas the day before the Nobel Prize and the
42:17
day after the Nobel Prize, right?
42:20
But it took the world had to catch up with to her rather than
42:25
it being the other way around. And it was the same way with
42:27
Galileo, right? Galileo followed the data and
42:31
eventually trusted it, and the world eventually caught up.
42:34
And so these folks, these scientists, and typically
42:39
they're scientific entrepreneurs in one form or another that
42:43
followed the data, stayed true to to themselves, preserved in a
42:50
world where they were, they were the opposite of the dogma.
42:55
And then one day they became dogma, if you will.
42:59
They became inevitable. Their science became obvious.
43:03
I find those journeys to be immensely interesting at at a
43:09
human level. And I know that obviously
43:12
there's a podcast where I I spoke quite a bit, but I would
43:16
like to think that I would love to have dinner with the two of
43:20
them or others. There are many others that are
43:23
representative as well and simply ask one question and then
43:28
just be quiet and listen, right? Because those, those journeys at
43:33
a human level are so in the things that they discovered that
43:37
made them know about laureates or made Galileo Galileo.
43:40
Of course, those are interesting.
43:42
But the human journey to preserver because if any of them
43:46
would have quit along the way, if any of them would have become
43:49
so frustrated that that they left the sciences because
43:54
because their work wouldn't being accepted, we wouldn't live
43:57
in the world that we live in today, right.
44:00
So that that understanding the journey and the the ability to
44:04
preserve veer in the face of folks saying that no, this can't
44:08
be right because, you know, you're not part of the Academy,
44:13
so to speak. You've done something that can't
44:17
be possible because you're not part of the Academy, right?
44:20
Those journeys are absolutely, absolutely fascinating to me.
44:24
And whether it's Doctor Carrico or it's Galileo, those folks I
44:30
find to be the most interesting people within the sciences.
44:34
So really exciting question and like I said, at least I'd like
44:37
to, I'd like to think I do quite a bit of, of listening to, to
44:41
that journey and the human side of, of changing the world.
44:46
Yeah, I'd love that question about the type of people as
44:50
well. Well, thanks so much for coming
44:52
on the podcast, Daniel. It's been a pleasure to have you
44:55
today. Thank you so much for for having
44:58
me and really, really appreciate the the, the question today.
45:02
This was really fun for for me as well and I hope the listeners
45:06
and those viewing it enjoy it. And can everyone remember to
45:10
like and subscribe?